[0001] This invention relates to circuits contained in and integrated to memory cell devices,
and particularly to a circuit for reading memory cells by a technique which utilizes
reference cells.
[0002] In order to fill the market's steadily growing demand for higher processing capacity,
computer and peripheral unit manufacturers are striving to provide faster access to
the data stored in memories of both the volatile and non-volatile types. In addition,
there exists a need for lower supply voltages in order to reduce power consumption,
while the development of memory cells of the multi-level type commands enhanced discriminating
capabilities and more accurate reading.
[0003] Thus, a pressing problem is to provide circuitry that can give access to stored information
in less time, and requires less power to do so, while ensuring highly reliable reading
of the stored data, to thereby fill the market's demands.
[0004] The problem of high-speed reading or sensing has been tackled in several ways.
[0005] Reading by reference cells involves the use of a non-volatile memory cell identical
with the memory cell being selected by the decode blocks; regardless of whether the
selected cell is written or not (i.e., has a high or low threshold), the reference
cell is always erased (low threshold). In this way, the read circuit is comparing
a low threshold cell (thereby drawing a nominal current typical of the physical characteristics
of the cell) with a cell whose threshold may be high or low, according to how it has
been programmed. The selected cell will draw substantially the same current as the
reference cell, if the threshold is low, and no current if the threshold is high.
[0006] A read circuit should supply the data output stages with either a high or a low logic
value, according to whether the threshold of the selected cell is the same as or other
than that of the reference cell.
[0007] A generic reading arrangement based on the use of a reference cell is showm in Figure
1. Its principle is that of having a generic selected cell (which may have a high
or a low threshold) compared with a constantly virgin cell which, as such, will always
be drawing the nominal current that characterizes it.
[0008] The reason why reading is performed by comparing a selected cell with a reference
cell is that it provides an always virgin cell capable of accommodating the process
variations of the array cells throughout the integrated circuit fabrication.
[0009] Reading by the reference cell technique actually is not effected by comparing any
selected cell with one reference cell, but rather by relating the bit line that contains
the cell to be read to the reference bit line. Thus, decoding will result in two cells
being selected in the same row of the array, namely the cell to be read and the reference
cell. Although the comparison is made with only one reference cell, the term reference
bit line is more appropriate.
[0010] The reference bit line does solve the problems from process variations. The reference
bit line approach, in fact, lowers the rate of variations from processing tolerances
in the physical and electrical parameters between cells as brought about by the memory
array being spread over a fairly large surface. Once the row (word line) is selected,
the reference bit line cell and the selected memory cell are allocated to the same
axis.
[0011] The advantage of the reference cell reading method over the differential cell reading
method (wherein a data is stored into two memory cells in its straight and negated
forms) resides in its reduced silicon area requirements.
[0012] As mentioned above, connected to the bit lines are, especially in high storage capacity
(4Mbit, 16 Mbit) devices, the drains of several thousands of memory cells. The capacitances
of the junctions associated with the drains of such cells add together into an overall
capacitive load of several picofarads. Again, the high bit line capacity is bound
to restrict the sensing speed of any reading scheme based on the amplification of
a voltage signal developed at the drain of the selected cell (voltage mode operation).
[0013] Therefore, current sensing appears preferable. The data discriminating circuitry
(sense amplifier) should have a low input impedance and be responsive to current,
rather than voltage, signals. The advantage of this method resides basically in its
low input impedance allowing the cell current to be injected into the sense amplifier
without the bit line voltage undergoing any significant change. In other words, the
current method enables the bit line capacitance to be allocated to a node which bears
little or no influence on the stored data sensing speed. Usually, before the reading
step itself, a "pre-charging" step is also carried out in order to attain optimum
voltage at the bit line node for performing the reading.
[0014] Reading a cell which has the same threshold voltage as the reference cell includes
a privileged reading and a difficult reading. Where the selected cell has a high threshold
voltage (and, therefore, no current flowing therethrough), the comparison with the
reference cell is easily carried out. With two cells having low thresholds, on the
other hand, the comparison becomes more difficult; the current flowing through the
two cells is the same, and the sensing circuitry is to detect this condition.
[0015] From the literature, many circuits for reading memory cells which are based on the
reference cell method are known, and their possible use is tied to the supply voltage,
whose value tends, as mentioned above, to decrease in today's applications.
[0016] The technical problem underlying the present invention is to provide a novel read
circuit for semiconductor memory cells, which is based on the reference cell method
and, additionally to including the best features of conventional circuits, exhibits
improved accuracy in discriminating data contained in a cell, while being specifically
suitable for use in multi-level cell storage devices.
[0017] This technical problem is solved by a read circuit as indicated above and defined
in the characterizing portions of the appended claims to this specification.
[0018] The features and advantages of a read circuit according to the invention will be
apparent from the following description of an embodiment given by way of example and
not of limitation with reference to the accompanying drawings.
[0019] In the drawings:
Figure 1 is a block diagram of a conventional memory cell read circuit using the reference
cell method technique;
Figure 2 is a diagram of a memory cell read circuit according to the invention; and
Figure 3 illustrates an application for reading multi-level cells, specifically four-level
cells.
[0020] Shown in Figure 2 is a read circuit for non-volatile memory cells which is based
on having capacitors charged through current mirrors, and in which the current of
the reference cell(s) and the array cell is mirrored onto the capacitors.
[0021] The principle underlying this reading technique is to arrange for capacitors to be
charged, after being discharged during the step of pre-charging the bit lines which
contain the reference cells and the memory cells in the array.
[0022] As the skilled ones in the art are aware, the reading operation includes two distinct
steps: a first step is the so-called pre-charging step whereby the reference bit line
and the memory array bit line are charged to a voltage of 1 volt in order to bring
the memory cell to be read to an optimum current supply condition; the second step
is the reading step proper and, according to the technique being used, may be carried
out in either a voltage or a current mode.
[0023] A circuit according to the invention utilizes this technique in that the array cell
and reference cell currents are mirrored onto two capacitors CR and CM which, according
to the known law of capacitor charging, begin to produce increasing voltages at nodes
outref and outsel (see Figure 1) depending on the current value delivered by the two
cells.
[0024] During the pre-charging step (signal ck "high"), the circuit will be inhibiting the
sensing structure, whose P-channel transistor M
1 and N-channel transistors M
PR and M
PM actually serve as switches to provide a ground connection for the nodes outref and
outsel at the same time as the currents from the bit lines are being mirrored during
the pre-charging step.
[0025] At the end of the pre-charging step, the reading step (signal ck low) begins; the
sensing circuit is enabled and transistors MSR and MSM, being enabled by MI, can mirror
the currents I
ref and I
cell from the respective reference and array branches. The two capacitors begin to be
charged and, according to the current value being delivered from the mirrors, their
voltage values will increase in different ways. Thus, if a larger current I
cell is delivered from the array side than the current I
ref from the reference side, the voltage at the node outsel will increase at a faster
rate than that at the node outref. This evolution of the signal allows for the use
of a fairly simple output structure AMP, out, such as a differential amplifier or
the like, which combines enhanced circuit compactness with correspondingly high accuracy
and speed of operation.
[0026] From the circuit standpoint, the size of the capacitors CR and CM is highly important;
in fact, too low a capacitance value would result in excessively fast, and hence inherently
inaccurate, charging due to sensitivity to variations in the integration process.
Conversely, high capacitance values would not only result in too slow a reading operation,
but also in unacceptably high silicon area occupation. Accordingly, a compromise must
be struck between speed and integration area requirements on the one side, and accuracy
on the other.
[0027] The concept of charging the capacitors at different slopes represents a simple and
fairly safe method, but it is a technique that, once the capacitors being charged
reach the asymptote, can provide no useful information for the next differential stage
AMP. In fact, without the N-channel transistors MLR and MLM, the circuit would evolve
to "saturation" of the nodes outref and outsel at the supply voltage, so that at the
end of the transient it would no longer be possible to discriminate the currents from
the reference cells and the array.
[0028] Thus, an important feature of the invention is the provision of the transistors MLR
and MLM in the circuit shown in Figure 2. From the time the trailing edge of the signal
ck initiates the reading step, the capacitors begin to be charged and the nodes outref
and outsel to rise in voltage; of course, the rising times are different, and one
of the two nodes will be first to attain a voltage (threshold voltage) driving the
transistor connected to the opposite branch to discharge the slower circuit counterpart.
[0029] Assuming, for instance, that the array cell is producing a larger current than the
reference cell, the capacitor CM will be charged at a faster rate than the capacitor
CR, to result in a faster increase of the voltage at the node outsel than at outref.
Upon the node outsel reaching the threshold voltage of the transistor MLR, the latter
will tend to go on and discharge the capacitor CR. Since CR already becomes charged
at a slower rate because of the smaller current mirrored by the reference cell, the
process will be further slowed by CR being discharged through MR.
[0030] In the meantime, however, Cm continues to be charged according to the capacitor charging
law, and consequently, the voltage at the node outsel also continues to rise; however,
this means increased "overdriving" of transistor MLR, which transistor will begin,
at a certain point, not only to interfere with the discharging of CR, but also to
discharge it back to ground level.
[0031] Thus, the two nodes in the steady-state will be at logic values of potentials Vdd
and gnd, with the big advantage that the next cascaded differential stage is able
to output the data in a stable manner without resorting to structures for storing
the reading performed and related read or strobe signals for driving them.
[0032] It should be emphasized that this type of sensing is based more on the capacitor
charging than the operation of the cross structure of the transistors MLR and MLM.
The sensing itself is effected during the charging. The transistors MLR and MLM serve
to keep the data stable and prevent the capacitor that is discharging at a slower
rate from attaining the charging asymptote.
Practical Applications
[0033] The proposed circuit can be used with two-level cells of the standard type as well
as with cells of the new multi-level type. With a two-level technique, the reference
is represented by a memory cell whence an intermediate current is flowing between
the maximum value delivered by the cell and zero; in this way, when the comparison
is effected between a cell which is delivering no current and the reference cell,
the latter will tend to charge the reference capacitor CR at a faster rate, whereas
the steady-state capacitor CM is in a discharged state. On the other hand, when the
array cell is virgin, it will tend to discharge the capacitor CM at a faster rate
to give rise to the opposite situation from that previously described.
[0034] For memory cells of the multi-level type, with four levels and a technique of the
parallel type [C. Calligaro, V. Daniele, E. Castaldi, A. Manstretta, N. Telecco, C.
Torelli, "Reading circuit for multilevel non-volatile memory cell devices", Pat. Dep.
SGS-Thomson, Europe (FR DE GB IT), No. 95830127.7 (1995)], three references are needed
for the read operation, and the cell current must be not only mirrored but also duplicated
thrice, so that it can be compared with the above-described references by three circuits
of the type of Figure 2.
[0035] Of course, the references should be programmed with suitable threshold values such
that the cells led to them can deliver currents having intermediate values to the
values that the generic array cell can take.
[0036] A practical example for a multi-level of four levels is the following:
- 1st cell level (virgin)
- 100 µA
- 2nd cell level
- 70 µA
- 3rd cell level
- 40 µA
- 4th cell level
- 0 µA
whereas for references:
- 1st reference
- 85 µA
- 2nd reference
- 55 µA
- 3rd reference
- 25 µA
[0037] In this way, all currents will be different and regardless of the cell being selected,
it will always locate between an upper and a lower reference. The capacitors will
therefore be charged always with different times, and this will cause the differential
stages connected to the nodes outref and outsel to output univocal conditions for
the cell being read.
[0038] Shown in Figure 3 is the circuit structure of this invention for this example of
a four-level reading.
[0039] The present sense circuit can also be used for multi-level readings of the serial
and serial dichotomic types [C. Calligaro, V. Daniele, R. Gastaldi, A. Manstretta,
C. Torelli, "Serial dichotomic method for sensing multiple levels non-volatile memory
cells, and sensing circuit acting such method"; C. Calligaro, V. Daniele, R. Gastaldi,
N. Telecco, A. Manstretta, C. Torelli, "Sensing circuit for serial dichotomic sensing
of multiple levels non-volatile memory cells", Pat. Dep. SGS-Thomson, Europe (FR DE
GB IT), No. 95830110.3 (1995)], although the parallel-type technique just described
is the most natural.
[0040] To summarize, the proposed circuit has the following novel features:
the use of charged capacitors as reading elements;
the use of transistors with their gates driven by output nodes to discharge the node
connected to the slowest-charged capacitor, so as to first slow down and then fully
discharge the slow capacitor.
[0041] Of course, the skilled ones in the art could apply changes and substitutions in the
inventive circuit described hereinabove.
1. A read circuit for semiconductor memory cells, comprising first and second active
elements (Msr,Hsm) coupled to a supply line (Vdd) via at least a first switch (M1), characterized in
that said first and second active elements are respectively connected, at a first
(outref) and a second (outsel) circuit node, respectively, to first (MLR) and second (MLN) threshold switches whereby the active elements are coupled to a ground (gnd), and
that said first and second circuit nodes are also connected to said ground through
first (CR) and second (CM) capacitive elements, respectively, each having a switch
(MPR,MPM) connected in parallel to the capacitive element.
2. A read circuit according to Claim 1, characterized in that each of the first (MLR) and second (MLM) threshold switches has a control terminal, with the control terminal of the first
threshold switch being connected to the second circuit node (outsel) and the control
terminal of the second threshold switch being connected to the first circuit node
(outref).
3. A read circuit according to either of Claims 1 and 2, characterized in that the first
and second active elements are respectively driven through first (MDR) and second
(MDM) input circuit elements, and the first and second circuit nodes (outref,outsel)
are connected to a differential output (out) stage (AMP).
4. A read circuit according to Claim 3, characterized in that the input circuit elements
are diode-connected transistors.
5. A read circuit according to any of Claims 1, 2, 3 and 4, characterized in that the
switches (MPR,MPM) connected in parallel to the first and second capacitive elements
(CR,CM) operate with phases which do not overlap those with which the first switch
(M1) operates.
6. A read circuit according to Claim 5, characterized in that the active elements and
switches are transistors of the MOS type and the capacitive elements are capacitors.
7. A storage device with semiconductor memory cells, characterized in that it comprises
at least one read circuit as claimed in any of Claims 1, 2, 3, 4, 5 and 6.
8. A storage device with memory cells and corresponding reference cells, characterized
in that it comprises at least one read circuit as claimed in any of Claims 1, 2, 3,
4, 5 and 6.
9. A storage device with memory cells and corresponding reference cells, characterized
in that it comprises at least one read circuit as claimed in any of Claims 1, 2, 3,
4, 5 and 6 having the input circuit elements coupled, by means of data transmission
lines, the one to at least one memory cell and the other to the corresponding reference
cell to said memory cell.
10. A read circuit for semiconductor memory cells of the n-level type, characterized in
that it comprises, for reading at least one of said cells, at least n-1 read circuits
as claimed in any of Claims 1, 2, 3, 4, 5 and 6.
11. A multi-level memory cell device, characterized in that it comprises a read circuit
as claimed in Claim 10.